Levels of evidence: study types and which papers to trust first
By Mubarak MutesasiraUpdated 11 min read
Short answer:
Levels of evidence rank study designs by how likely they are to answer a question without bias. For treatment questions the usual order, from strongest, is: systematic review of randomized trials, randomized controlled trial, cohort study, case-control study and case series, then expert opinion and mechanism. The right top design changes with the question, and a well-run study can outrank a poor one above it.
This guide is for students, clinicians and researchers deciding which papers to read first and how much weight to give each. It is the pillar of our series on finding and checking sources.
What are the levels of evidence?
Levels of evidence are a ranking of study designs by how well each design protects its answer from bias and chance, for a given type of question. They are a shortcut: when you have hundreds of results, they tell you which box to open first.
The most widely used scheme is the Oxford Centre for Evidence-Based Medicine (OCEBM) Levels of Evidence, revised in 2011. Its introductory document describes it as "a hierarchy of the likely best evidence" designed as a short-cut for busy clinicians, researchers and patients. [1] For the question "Does this intervention help?" the levels are: [1][2]
| Level | Treatment benefit: design |
|---|---|
| 1 | Systematic review of randomized trials or n-of-1 trials |
| 2 | Randomized trial, or observational study with a dramatic effect |
| 3 | Non-randomized controlled cohort or follow-up study |
| 4 | Case series, case-control studies, or historically controlled studies |
| 5 | Mechanism-based reasoning |
The levels are "NOT intended to provide you with a definitive judgment about the quality of evidence," and they do not give a recommendation. [1] A level tells you where to start reading, not what to conclude.
What are the main types of research studies?
Research studies are experimental (the researcher assigns the exposure), observational (the researcher watches what happens) or secondary (studies of studies).
| Study type | What the researchers do | Best for | Main weakness |
|---|---|---|---|
| Meta-analysis | Pool the numerical results of several studies into one estimate | A precise overall effect size | Only as good as the studies pooled; mixing unlike studies |
| Systematic review | Search, select and appraise all studies on a question by a written protocol | A complete, reproducible summary | Slow; limited by the quality of included studies |
| Randomized controlled trial (RCT) | Randomly assign people to treatment or control and compare outcomes | Whether a treatment causes a benefit | Costly; short; strict entry criteria limit who results apply to |
| Cohort study | Follow a group with and without an exposure over time | Prognosis, causes, rare harms, long-term outcomes | Confounding: exposed and unexposed groups differ in other ways |
| Case-control study | Compare people with an outcome (cases) to people without it (controls), looking back at exposures | Rare diseases and outcomes | Recall bias; choosing fair controls is hard |
| Cross-sectional study | Measure exposure and outcome at one point in time | Prevalence; accuracy of a diagnostic test | Cannot show which came first |
| Case report or case series | Describe one or several patients | New diseases, unusual side effects | No comparison group |
| Qualitative study | Interviews, focus groups, observation; analyze meaning | Experiences, reasons, barriers | Not built to measure how often or how much |
| Expert opinion, mechanism | Reasoning from physiology, lab work or experience | Generating hypotheses | Plausible mechanisms often fail in people |
A randomized controlled trial is a study that assigns participants to groups by chance, so known and unknown differences are balanced and a difference in outcome can be attributed to the intervention.
A cohort study is an observational study that follows people with and without an exposure over time and compares how often an outcome occurs.
A case-control study is an observational study that compares people who have an outcome with a suitable control group who do not, and looks back for differences in past exposure. [3]
A systematic review is a review that answers a defined question by finding, selecting and appraising all eligible studies with explicit, reproducible methods. A meta-analysis combines their numbers. See what a systematic review is.
What is the evidence pyramid?
The evidence pyramid is a diagram of the hierarchy of evidence that stacks study designs from the weakest and most plentiful at the base to the strongest at the top. From the top:
- Systematic reviews and meta-analyses
- Randomized controlled trials
- Cohort studies
- Case-control studies
- Cross-sectional studies, case series and case reports
- Expert opinion, mechanism, bench research
What are the limits of the evidence pyramid?
The pyramid is useful as a reading order and misleading as a verdict. Four limits matter:
- Design is not quality. A small RCT with 40% dropout can be weaker than a large, careful cohort study. The OCEBM levels say a level can be graded down for poor study quality, imprecision, inconsistency between studies, indirectness (the study's population or outcome doesn't match your question) or a very small absolute effect, and graded up for a large effect. [2]
- Dramatic effects count. An observational study with a dramatic effect can be stronger evidence than an inconclusive review of a few studies. [1]
- Reviews are a lens, not a layer. Murad and colleagues' 2016 revised pyramid makes the lines between designs wavy, since certainty moves with quality, and takes systematic reviews off the top to use them as a lens for viewing the other studies. [4]
- The pyramid is built for treatment questions. For other questions the order changes, as the next section shows.
GRADE is a system that rates the certainty of a body of evidence for one outcome as high, moderate, low or very low, starting from the study design and then moving down or up for specific reasons. [5] Guideline panels and Cochrane reviews use it.
What is the highest level of evidence for each type of question?
The highest level of evidence is a systematic review of the best primary design for the question, and that design differs by question. The OCEBM table has a row for each question type [2]; the last row below is outside it.
| Your question | Example | Best primary design |
|---|---|---|
| Does this treatment help? | Does cognitive behavioral therapy reduce insomnia? | Randomized controlled trial |
| How common is it? | How many nurses report burnout in this region? | Current, local random-sample survey (cross-sectional) |
| Is this test accurate? | How well does a rapid test detect strep throat? | Cross-sectional study comparing the test with a reference standard in consecutive patients |
| What will happen without treatment? (prognosis) | What share of people with a first stroke have a second within five years? | Inception cohort study |
| What are the rare harms? | Does a drug raise the risk of a rare liver injury? | Randomized trials if large enough; in practice often large cohort or case-control studies |
| How do people experience it? | Why do patients stop taking a medicine? | Qualitative study; qualitative evidence synthesis |
An inception cohort is a group of people assembled at the same early point in their condition, such as diagnosis, then followed forward.
Is a systematic review level 1 evidence?
A systematic review of randomized trials is level 1 evidence for treatment questions. A systematic review of case series is not. The level of a review depends on the designs it includes and how well they were done.
Check three things before you treat a review as level 1:
- What designs did it include? "RCTs only" and "any study design" give very different reviews.
- Was the search systematic? It should name the databases, dates and search strings, and report a flow of records (often a PRISMA diagram).
- Did it rate certainty? A GRADE "summary of findings" table shows whether the authors rated the evidence high or very low.
What level of evidence is a cohort study?
A non-randomized controlled cohort study is level 3 evidence for treatment questions in the OCEBM scheme. [2] For prognosis, a systematic review of inception cohorts is level 1.
How do you identify a study's type from its abstract?
Most abstracts name the design in the title or methods. When they don't, look for these signals:
- "Randomly assigned", "allocated to": an RCT.
- "Followed for", "incidence", "hazard ratio": a cohort study.
- "Cases", "controls", "odds ratio", with past exposures: a case-control study.
- "Survey", "prevalence", "at a single time point": a cross-sectional study.
- "Interviews", "focus groups", "thematic analysis": a qualitative study.
- "We searched", "pooled": a systematic review or meta-analysis.
Database labels can be wrong or missing: PubMed's article-type filters rely mainly on the publication type field [6]. Check the methods.
How do you filter a search by study type?
Filter by study type so the strongest designs come first. For a treatment question:
- Search for systematic reviews first. In PubMed, add the Systematic Review or Meta-Analysis article-type filter, or the tags
systematic review[pt]andmeta-analysis[pt]. [6] - Check the newest review's search date. If it searched up to 2021, you need trials published since.
- Search for randomized trials after that date. Use the Randomized Controlled Trial filter (
randomized controlled trial[pt]) with a date range. - Add observational studies only where trials can't answer: long-term outcomes, rare harms, groups excluded from trials.
See how to search PubMed for the full syntax.
A worked example
This is an illustrative example with invented figures. A nursing student asks: "Does supervised exercise reduce depressive symptoms in adults over 65?"
- She filters for systematic reviews and finds four. The newest searched to March 2023 and pooled 18 RCTs with 1,520 participants, rating certainty as moderate.
- She searches for RCTs from April 2023 and finds two new trials of 120 and 210 participants: 330 more people, about a fifth of the pooled sample.
- She skips the 40 cross-sectional studies: they show active older adults report fewer symptoms, but not that exercise caused it.
Her answer cites the review, then the two newer trials, with the review's certainty rating. Evidence like this informs care only after a qualified professional has reviewed it for the specific patient.
How do you find and rank evidence in AI for Research?
AI for Research labels each paper with its study type and lets you filter by it.
- Find Papers runs one search across the sources you choose, including PubMed and Europe PMC. Filter by study type (from meta-analysis and RCT to cohort, case-control, cross-sectional, qualitative and case report), year, journal quality tier, citations and open access. Retracted papers are left out by default. Searching and filtering cost 0 credits.
- Study type labels come from PubMed, Europe PMC and Semantic Scholar first, then from the title.
- Deep Research answers a question with citations and a consensus line counted from per-study labels, such as "2 of 4 studies that test the question support it". Choose the Medical source to search PubMed and Europe PMC. A Summary run costs 1 credit; Deep costs 5.
- Claim check shows whether studies support up to 10 claims from a paragraph, with the papers, for 2 credits.
Limits: study-type labels can be wrong, especially those taken from titles. The tool doesn't apply GRADE or OCEBM levels for you. Where no open full text exists, Deep Research reads the abstract only. The free plan includes 45 credits a month; see pricing.
FAQ
What are the levels of evidence?
Levels of evidence rank study designs by how likely they are to give a trustworthy answer to a given type of question. For treatment questions, the Oxford CEBM levels put systematic reviews of randomized trials at level 1, single randomized trials at level 2, controlled cohort studies at level 3, case-control studies and case series at level 4, and mechanism-based reasoning at level 5.
What is the highest level of evidence?
For questions about whether a treatment works, the highest level is a systematic review of randomized controlled trials, ideally with a meta-analysis. For other questions the top design changes: a systematic review of inception cohort studies for prognosis, and of cross-sectional accuracy studies for diagnosis. A poorly done review of small, biased trials can still be weaker evidence than one large, well-run trial.
Is a systematic review level 1 evidence?
A systematic review of randomized trials is level 1 evidence for treatment questions in the Oxford CEBM scheme. A systematic review of weaker designs, such as case series, is not automatically level 1. The level depends on the studies inside the review, and a review can be graded down for poor quality, inconsistent results, imprecision or indirectness.
What level of evidence is a cohort study?
For treatment questions, a non-randomized controlled cohort study is level 3 in the Oxford CEBM scheme, below randomized trials. For prognosis questions, an inception cohort study is the best single design, and a systematic review of inception cohorts is level 1. A cohort study is also often the best available design for harms that randomized trials are too small or too short to detect.
Where do qualitative studies fit in the hierarchy of evidence?
Qualitative studies sit outside the classic treatment hierarchy because they answer different questions: how people experience an illness, why a program fails, what patients value. For those questions they are the right design, and a qualitative evidence synthesis is the strongest form. Judge them with qualitative appraisal tools, not by their position on a treatment pyramid.
Next steps
- Put the levels to work in a database: how to search PubMed with MeSH terms and filters.
- Find the studies a good review missed with citation snowballing.
- Weigh a specific statement against the studies with how to check a claim against the research.
- Keep what you find in order with how to organize research papers.
Sources
- Howick, J., Chalmers, I., Glasziou, P., et al. The 2011 Oxford CEBM Levels of Evidence (Introductory Document). Oxford Centre for Evidence-Based Medicine. https://www.cebm.ox.ac.uk/resources/levels-of-evidence/levels-of-evidence-introductory-document
- OCEBM Levels of Evidence Working Group. The Oxford Levels of Evidence 2. Oxford Centre for Evidence-Based Medicine. https://www.cebm.ox.ac.uk/resources/levels-of-evidence/ocebm-levels-of-evidence
- Oxford Centre for Evidence-Based Medicine. Glossary. https://www.cebm.ox.ac.uk/resources/ebm-tools/glossary
- Murad, M. H., Asi, N., Alsawas, M., & Alahdab, F. (2016). New evidence pyramid. Evidence-Based Medicine, 21(4), 125–127. https://doi.org/10.1136/ebmed-2016-110401
- Guyatt, G. H., Oxman, A. D., Vist, G. E., Kunz, R., Falck-Ytter, Y., Alonso-Coello, P., & Schünemann, H. J. (2008). GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ, 336(7650), 924–926. https://doi.org/10.1136/bmj.39489.470347.AD
- National Library of Medicine. PubMed User Guide. https://pubmed.ncbi.nlm.nih.gov/help/